What was a Photo? The Week in Review
Google’s Reimagine tool unleashes hell. No, really.
It’s been a whirlwind of a week following the reveal of Google’s new AI-powered photo editors. So much so that we both decided to write our own separate pieces. Happy Friday, y’all.
Reimagine that.
Shortly after we published this week's essay on photo surrealism, The Verge posted its review of Google’s new lineup of Pixel 9 phones. The Verge's Chris Welch also posted an incredible set of pictures on Threads he created using Google's "reimagine" feature (a tool within the Magic Editor) and... well, the photos speak for themselves.
Welch managed to add car wrecks, a helicopter crash, drug paraphernalia around a photo of a friend, bombs in subway stations and even corpses covered in a bloody sheet in a crash scene to his photos within a matter of seconds just by typing a prompt.
As I wrote in my essay, all of these creative photo editing techniques have been available to photographers for decades. However, up to now, this level of retouching was time-consuming, difficult, and required powerful and expensive computers.
Google's new AI-powered editing tools Add Me and Magic Editor, especially the "reimagine" feature, empower anyone to apply complex and very realistic retouching and compositing techniques—seemingly without any guardrails or restriction—to any photo they want within seconds, and using only their phone! It's not so much what "reimagine" can do but the speed and ease with which it can do it.
But what I find genuinely alarming about this is that there is nothing to alert the viewer that the photo has been altered using AI (except deep down in the metadata) and that the image, or at least certain elements of if, isn't real. This seems incredibly irresponsible on Google's part, especially less than 73 days away from the US election.
I previously wrote that putting these powerful AI photo editing tools in phones "is a genuine concern."
After seeing the photos in The Verge's review and the photos taken by Chris Welch, I now think that was a monumental understatement on my part. To quote Sarah Jeong from her recent essay on The Verge, “We are fucked.” —James
…Welp!
It's a question of guardrails, ultimately. When Grok's generative AI tools launched last week, what surprised people the most was how easy it was to find the loopholes and create anything you want, up to and including Mickey Mouse holding a gun standing over graphic images of dead children. This week, as they say, was a hold-my-beer moment.
While I haven't seen quite the level of gore come out of Pixel-created imagery, there are still plenty of clear examples of famous cartoon characters shooting one another or dressing up in Nazi uniforms—images that already existed all over certain corners of the internet but until now haven't been so easy to create by writing prompts in a smartphone.
Compared with reimagine, though, all of this is relatively tame. Few people really think Barney the Dinosaur would actually shoot Elmo with an assault rifle. It's a far different story once you start adding convincing generative imagery to amateur smartphone photography. When I saw Welch’s photo of a NYC subway with a convincing-at-first glance bomb, I immediately thought about how many of these we'll be seeing in the months and years to come, and all the ways in which a "troll" image will cause true panic. Absent guardrails on what can be generated—and absent effective detection tools to help us understand truth from fiction—we're in for a very, very messy time. How long until the first AI-manipulated photo shuts down a subway station or public event? What unedited photos of vital importance will be decried as fake because the subject matter is too upsetting to want to believe otherwise?
There's a good chance this fear is going to feel abstract for a time. Right now these tools are really only available for the power users and early adopters. Eventually, though, the newest model becomes the discount model. We're just a few years and a few phone generations from everyone being able to quickly and convincingly manipulate smartphone images. Which means many years of people both falling for convincing fakes and dismissing unaltered photos. There’s still time to figure out these guardrails and test kits before the floodgates open. I’m not exactly hopeful, but I hope to be wrong about that.
And yeah, I also now get why Apple’s been holding back on adding its image generation features to the iOS beta. —Ross
What else we’ve read
Condé Nast signs deal with OpenAI (Wired)
Condé Nast now joins the list of other media companies such as Vox Media, The Financial Times, The Atlantic, Axel Springer (Business Insider and Politico), News Corp and Dotdash Meredith who have now made licensing deals with OpenAI—The New York Times is following a different strategy and is (currently) suing OpenAI for billions.
The details of the Condé’s agreement with OpenAI haven’t been disclosed but Damon Beres at called these deals a “Devil’s Bargain” when he wrote about The Atlantic’s agreement. Others have compared these negotiations between the media and the AI companies as akin to a hostage situation. But, the fact that OpenAI is now actively looking to enter the search market makes these agreements seem a little less like slamming the door after the horse has bolted. Regardless, media companies still seem to have a very weak hand in these negotiations. —James
Pain at Penske (Status)
Coming on the back of the news about Condé Nast’s deal with OpenAI, Oliver Darcy reports that Penske Media “has quietly laid off staffers in piecemeal fashion over the last several weeks, raising questions about the health of the powerhouse publisher.”
It seems that Penske is joining a number of other media companies over recent months who have let people go in dribs and drabs seemingly to avoid negative headlines. But as Darcey points out, “it's notable for Penske to lay off staffers, given that his portfolio of outlets has historically proven to be quite lucrative.” —James
“We’re never going there. Creativity is made, not generated.”
This week, Procreate and its CEO James Cuda took a strong stance on generative AI:
Generative AI is ripping the humanity out of things. Built on a foundation of theft, the technology is steering us toward a barren future. We think machine learning is a compelling technology with a lot of merit, but the path generative AI is on is wrong for us.
While we at MBH4H are generally curious about the potential of generative AI existing alongside established creative outlets, we also think it’s very stupid when companies try to fit AI in everywhere. Photoshop with generative tools? Sure, that makes sense. One of the world’s best drawing apps? Yeah no. Keep GenAI away, please and thank you. —Ross
What we’ve watched
The Handmaid’s Tale (Hulu)
I am now one episode away from the finale of season five of The Handmaid’s Tale (the sixth and final season is slated to air in 2025) and I genuinely can’t believe that I waited this long to watch this show. I never finished Season 1 when it first aired in 2017, finding it slow and disturbing. Watching it now in 2024 I find it slow, disturbing and utterly compelling. Binging through five seasons over a matter of weeks has been like watching the director's cut of The Handmaid’s Tale movie.
Much has been written about how good Elizabeth Moss is in this show (praise that is richly deserved I might add). But the entire cast is extraordinary, particularly Yvonne Strahovski who plays Serena Joy Waterford with such an air of psychopathic tragic sadness.
As we head towards a US election with women’s rights under direct threat from Project 2025 and a Republican Vice Presidential candidate openly denigrating women who choose not to have children, I can’t but help think that I didn’t so much miss the Handmaid’s Tale the first time round as wait to watch it at the perfect time. —James
What we’ve played
Slay the Spire (Mega Crit)
This is the closest I have to comfort food in gaming form. In Slay the Spire, you start with a small and simple deck of cards that you use to attack and defend yourself in a series of escalating encounters as you work your way up the eponymous tower. The more you win, the more cards you get to choose from and upgrade your deck. Lose even one battle, however, and you have to start over. The game is entirely turn-based and you know what the enemy is planning to do at the start of each round, so despite the stakes, it can be a very chill experience, giving you time to think through every step. I'm pretty sure I've hit over 500 hours in this game across both Switch, PC, and mobile. This is a cry for help. —Ross
What we’ve listened to
Folklore - Taylor Swift (Republic Records)
Walt Mosberg’s reviews of Taylor Swift are a joy to read and I have to credit his genuine and deep admiration for Swift as an artist as my inspiration to listen to more of her music. I have turned into a little bit of a Swifty as a result—though admittedly not on the same level as Walt. I find her utterly inspiring and the album I keep coming back to is Folklore. I love the simplicity and the almost “unplugged” sound of the album. Somehow I found myself inspired to listen to it again after spending all of last week listening to Fred again’s Tiny Desk concert on repeat. Fred again to Taylor Swift may not seem the most logical of leaps but it makes all the sense in the world to me. —James
What we’ve loved
This week's DNC in Chicago generated a ton of photographs and some great memes. But perhaps the single best photograph from the entire convention was taken on the last night by Todd Heisler for The New York Times.
Whatever happens over the next 73 days before the election, I can guarantee we’ll see this photo used over and over and over, deservedly so. It is perfect. —James